feat(community): implement Leiden community detection algorithm#401
Open
gkorland wants to merge 3 commits intoGraphBLAS:v1.3.xfrom
Open
feat(community): implement Leiden community detection algorithm#401gkorland wants to merge 3 commits intoGraphBLAS:v1.3.xfrom
gkorland wants to merge 3 commits intoGraphBLAS:v1.3.xfrom
Conversation
Add experimental implementation of the Leiden algorithm (Traag et al.,
Scientific Reports 9:5233, 2019) — the successor to Louvain that
guarantees well-connected communities.
New files:
- experimental/algorithm/LAGraph_Leiden.c: two-phase implementation
* Phase 1 (Local Move): greedy node-to-community assignment maximising
modularity gain score(i→c) = T[c] - k[i]*k_comm[c]/m
* Phase 2 (Refinement): Leiden's key addition — restart each node in a
singleton sub-community and restrict moves to within the Phase-1
parent community, guaranteeing well-connectedness
* Phase 3 (Aggregation): documented as TODO for multi-level extension
* Output: GrB_Vector of INT64 community labels 0..K-1
- experimental/test/test_leiden.c: acutest test using karate.mtx;
verifies all nodes labelled, labels in range, and Q > 0
Updated files:
- include/LAGraphX.h: add LAGRAPHX_PUBLIC declaration for LAGraph_Leiden
Algorithm is auto-discovered by experimental/CMakeLists.txt glob.
Tested on karate.mtx (Q ≈ 0.238, 34 nodes, all checks pass).
Ref: https://doi.org/10.1038/s41598-019-41695-z
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Wrap Phase 1 + Phase 2 in an outer aggregation loop that iterates until no further coarsening occurs - Build aggregate graph A_agg = S^T * A_cur * S after each level using GrB_mxm with GrB_PLUS_TIMES_SEMIRING_FP64 - Carry forward Phase-1 partition as initial partition for each aggregate level (Leiden invariant: init_comm[r] = Phase-1 parent of refined community r) rather than reinitialising from singletons - Skip self-loops (j == i) in both Phase 1 and Phase 2 move scoring; A_agg has diagonal entries representing intra-community weight - Recreate GrB_Vector k_vec and v per level to match current n_cur - Track o_comm[i] (original node → final community) across all levels via composition o_comm[i] = c_ref[o_comm[i]] after each relabeling - m (total edge weight / 2) is computed once from G->A and held constant; invariant under the S^T*A*S aggregation - Modularity Q on karate.mtx improves from ~0.24 (single-level) to ~0.34 (multi-level); update test threshold to Q > 0.3 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
…s review Code review on PR GraphBLAS#401 identified three issues; this commit addresses all three. - fix: change modularity-gain penalty from k_i*k_comm[c]/m to k_i*k_comm[c]/(2m) in both Phase 1 and Phase 2. The standard Blondel modularity-gain formula uses 2m as the denominator; using m doubled the penalty term and biased the algorithm toward over-fragmenting communities. Q on karate.mtx improves from 0.345 to 0.419, matching published Louvain/Leiden benchmarks (0.37-0.42). Updated test threshold to Q > 0.37. - fix: hoist A_new declaration to function scope and add it to LG_FREE_WORK so it is freed if the second mxm call returns an error. Previously A_new could leak on the error path. - docs: clarify in the header comment that the refinement phase is the greedy variant (not the randomized procedure from Traag et al. 2019). The seed parameter is reserved for a future randomized refinement step; added an explicit (void) seed cast to silence the unused-parameter reading. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Member
|
Thanks! Can you move the PR to the 1.3.x branch? That's the latest development version. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Implements the Leiden community detection algorithm (Traag et al., 2019) as an experimental LAGraph algorithm. Leiden is the successor to Louvain and guarantees well-connected communities by adding a Refinement phase between the Local-Move and Aggregation phases.
Changes
Testing
Memory / Performance Impact
Related Issues
Traag, V.A., Waltman, L. & van Eck, N.J. (2019). From Louvain to Leiden: guaranteeing well-connected communities. Scientific Reports 9, 5233.